On Unified Prompt Tuning for Request Quality Assurance in Public Code Review

公共代码审查（PCR）可以通过软件问答（SQA）社区实现，这有助于高效传播知识。目前的方法主要关注审阅者的角度，包括找到能力强的审阅者、预测评论质量以及推荐/生成审阅评论。我们的直觉是，满足审阅需求请求可以增加它们的可见度，从而为更好的审阅响应打下基础。为此，我们提出了一个统一的框架UniPCR，用遮蔽语言模型（MLM）完成基于开发人员的请求质量保证（即预测请求必要性和推荐标签子任务）。具体而言，我们通过以下两种方式重新构建了两个子任务：1）文本提示调整，使用硬提示构建提示模板，将两个子任务转化为MLM；2）代码前缀调整，使用软提示优化生成的连续向量的一个小段作为代码表示的前缀。对2011-2022年公共代码审查数据集的实验结果表明，我们的UniPCR框架适应了这两个子任务，并且在请求质量保证方面的准确性方面优于基于状态-最新方法的可比较结果。这些结论突出了我们的统一框架从开发者的角度在公共代码审查中的有效性。

UniPCR framework aims to complete developer-based request quality assurance in public code review, including predicting request necessity and recommending tags subtask. The problem is to improve the visibility of review necessity requests and obtain better review responses.

UniPCR framework reformulates the two subtasks via text prompt tuning and code prefix tuning under a Masked Language Model (MLM), which outperforms comparable accuracy-based results with state-of-the-art methods for request quality assurance.

The experimental results on the Public Code Review dataset for the time span 2011-2022 demonstrate the effectiveness of UniPCR framework from the developer's perspective. The framework adapts to the two subtasks and achieves better performance. The paper also highlights the importance of satisfying review necessity requests to increase their visibility and improve review responses.

Related work includes previous studies on Public Code Review and Software Question Answering (SQA) community, such as 'A Study of Public Code Review Practices in Open Source Projects' and 'Software Question Answering: A Survey'.